doi: 10.3389/fphys.2016.00071
Edited by:
Johnny Padulo, Università eCampus, Italy Reviewed by:
Pantelis Theodoros Nikolaidis, Hellenic Army Academy, Greece Alessandro Moura Zagatto, Uni Estadual Paulista-UNESP, Brazil
*Correspondence:
Billy Sperlich billy.sperlich@uni-wuerzburg.de
Specialty section:
This article was submitted to Exercise Physiology, a section of the journal Frontiers in Physiology Received:05 January 2016 Accepted:15 February 2016 Published:09 March 2016 Citation:
Düking P, Hotho A, Holmberg H-C, Fuss FK and Sperlich B (2016) Comparison of Non-Invasive Individual Monitoring of the Training and Health of Athletes with Commercially Available Wearable Technologies.
Front. Physiol. 7:71.
doi: 10.3389/fphys.2016.00071
Comparison of Non-Invasive
Individual Monitoring of the Training and Health of Athletes with
Commercially Available Wearable Technologies
Peter Düking1, Andreas Hotho2, Hans-Christer Holmberg3, 4, Franz Konstantin Fuss5and Billy Sperlich1*
1Integrative and Experimental Training Science, Department of Sports Science, Institute for Sport Sciences,
Julius-Maximilians University Würzburg, Würzburg, Germany,2Data Mining and Information Retrieval Group, Computer Science VI, Artificial Intelligence and Applied Computer Science, Julius-Maximilians University Würzburg, Würzburg, Germany,3Department of Health Sciences, Swedish Winter Sports Research Centre, Mid Sweden University, Östersund, Sweden,4School of Sport Sciences, UiT The Arctic University of Norway, Tromsø, Norway,5Department of Mechanical and Automotive Engineering, School of Engineering, RMIT University, Melbourne, Australia
Athletes adapt their training daily to optimize performance, as well as avoid fatigue, overtraining and other undesirable effects on their health. To optimize training load, each athlete must take his/her own personal objective and subjective characteristics into consideration and an increasing number of wearable technologies (wearables) provide convenient monitoring of various parameters. Accordingly, it is important to help athletes decide which parameters are of primary interest and which wearables can monitor these parameters most effectively. Here, we discuss the wearable technologies available for non-invasive monitoring of various parameters concerning an athlete’s training and health. On the basis of these considerations, we suggest directions for future development. Furthermore, we propose that a combination of several wearables is most effective for accessing all relevant parameters, disturbing the athlete as little as possible, and optimizing performance and promoting health.
Keywords: wearable technologies, performance parameters, health monitoring, performance monitoring, sports technology
INTRODUCTION
The survey of fitness trends world-wide published in December 2015 (Thompson, 2015) indicates that in 2016 for the first time, wearable technology will become the most popular and leading trend, with the wearable technology market approaching $6 billion dollars. Other trends in fitness, such as body weight training (ranked second in 2016) and high-intensity interval training (ranked sixth) have changed by no more than one place in ranking compared to 2015 (Thompson, 2014).
In contrast, in 2015 wearable technology was not ranked at all, probably because it was not even included in the survey.
Adaptation of training is highly individual (Bouchard et al., 1986), depending in part on the balance between exercise and recovery. A suboptimal training load can result in stagnation or de-adaptation, whereas overly intense and/or prolonged training may lead to chronic fatigue,
overreaching or overtraining, and negative health effects (Borresen and Lambert, 2009; Buchheit, 2014; Halson, 2014a). In this context, continuous (non-invasive) monitoring of biological and psychological markers might be helpful (Halson, 2014a), and since wearables offer the opportunity to measure different markers conveniently, they provide a promising approach.
Wearables are lightweight, sensor-based devices which are worn close to and/or on the surface of the skin, where they detect, analyze, and transmit information concerning several internal and/or external variables to an external device and provide in some cases immediate biofeedback to the athlete. However, the variety of such wearables already available is overwhelming and it is not clear which one(s) may be best for monitoring training and health.
Accordingly, our present aims are threefold: (a) to briefly summarize (non-invasive) parameters that are of potential value in assessing an athlete’s training and health; (b) to provide a brief overview of the individual wearables presently available and the parameters they monitor; and (c) to highlight current gaps in our knowledge in order to help direct both future scientific studies and the development of commercial wearables.
CANDIDATE VARIABLES FOR
(NON-INVASIVE) MONITORING OF AN ATHLETE’S TRAINING AND HEALTH Monitoring Training Status
Monitoring of an athlete’s training status must take into consideration the external load applied (i.e., the work completed) in relationship to the individual’s response to this load, and a recent review has nicely summarized the various internal and external parameters of potential interest in this context (Halson, 2014a). These parameters and their response to training are highly complex and it is beyond the scope of the present review to discuss them in detail. We simply outline key parameters briefly and refer readers interested in more information to other publications (Borresen and Lambert, 2009;
Halson, 2014a).
The external load is usually reflected in parameters such as distance (e.g., when running), velocity (e.g., of running), the duration and frequency of training sessions, etc. (Halson, 2014a). In addition, environmental conditions such as altitude, temperature, and relative humidity can influence the external load (Hargreaves, 2008; Mazzeo, 2008; Drust and Waterhouse, 2010; Maughan et al., 2012; Born et al., 2014) and should therefore be monitored as well.
Among the great variety of relevant internal parameters, some can only be monitored with sophisticated instruments and/or are invasive (e.g., blood analysis) and thereby impractical for daily use (Halson, 2014a). From a practical point of view, monitoring of internal parameters should not only be non-invasive, but also efficiently provide daily simple, yet scientifically trustworthy feedback designed to improve performance and maintain health.
Examples include heart rate (HR) during exercise (HRex), as well as recovery (HRR) and variability (HRV) of HR (Achten and Jeukendrup, 2003; Buchheit, 2014; Halson, 2014a).
The HRR is defined as the rate of decline in HR following termination of exercise, which is regulated by the autonomic nervous system and thereby provides information concerning sympathetic and parasympathetic activity (Daanen et al., 2012).
In general, the more rapid the HRR, the better the fitness (Daanen et al., 2012; Buchheit, 2014). However, since in trained endurance athletes a period of functional overreaching also appears to be associated with more rapid HRR, this parameter must be evaluated in the context of the training schedule (Aubry et al., 2015).
The HRV, defined as the time that elapses between two heart beats (Achten and Jeukendrup, 2003), can reveal alterations in the autonomous nervous system of the heart (Buchheit, 2014). Even though its applicability is debated (Plews et al., 2013; Halson, 2014a), when assessed longitudinally and at specific time-points (during the night or immediately after waking-up) HRV can help reveal an athlete’s training and health status (Plews et al., 2013, 2014; Buchheit, 2014).
In addition to these parameters related to the heart, elevated neuromuscular fatigue (defined as a reduction in force generation due either to central and/or peripheral factors) has been associated with symptoms of overtraining and should be monitored frequently (Fowles, 2006; Cormack et al., 2008;
Buchheit, 2014).
Moreover, different lactate thresholds are commonly used to determine an athlete’s internal loading and can be used to access the results of training interventions (Bellotti et al., 2013; Halson, 2014a). Consequently, in connection with monitoring an athlete’s training status, blood levels of lactate should also be taken into consideration.
Monitoring Health Status
Even though the parameters described above are related to those discussed in this section, we highlight here those that provide deeper insight into the training related health status of athletes (Speedy et al., 2001; Halson, 2014b; Saw et al., 2015).
Assessment of hydration status (which is influenced both by the extent of sweating and drinking behavior) is necessary, since dehydration can impair performance and, moreover, is associated with several deleterious health consequences, including heat strokes (Sawka et al., 2007). At the same time, overdrinking can result in hyponatremia and subsequent fatigue, confusion, coma, and even death (Speedy et al., 2001). Consequently, monitoring both fluid loss by sweat and fluid intake is of considerable importance.
When exercising in extreme environments, the athlete’s core temperature can exceed 40◦C (hyperthermia) or be less than 35◦ C (hypothermia), which can lead to several kinds of injuries and even threaten life (Armstrong et al., 2007; Fudge et al., 2015).
Ultraviolet (UV) radiation can damage DNA (Cadet et al., 2005) and is a major risk factor for melanoma and other forms of skin cancer (Moehrle, 2008). Consequently, athletes exercising outdoors should monitor their exposure to sunlight, both direct and reflected.
An alteration in the athlete’s arterial blood oxygenation (SpO2) may explain decrements in performance (Siegler et al., 2007),
especially at altitudes where this value is lowered, and may also help to predict acute mountain sickness (Basnyat, 2014).
The quality and quantity of sleep, especially slow-wave sleep during which growth hormones are secreted, are important for recovery, performance, and health and should also be monitored (Halson, 2014b). Impaired sleep disrupts cognitive and immune functions, enhances daytime sleepiness, and reduces overall performance (Leeder et al., 2012; Halson, 2014b).
Subjective parameters, such as mood disturbances or perceived stress and inadequate recovery, can be assessed with different questionnaires that actually appear to provide a more sensitive and consistent evaluation of an athlete’s well-being and training load than objective markers (Saw et al., 2015).
Accordingly, such questionnaires should be applied with confidence in daily practice (Saw et al., 2015).
WEARABLE TECHNOLOGIES DESIGNED FOR INDIVIDUAL CONSUMERS
To evaluate how wearables may assist in monitoring an athlete’s training and health, the technologies involved and their abilities to detect specific parameters must be understood.
Several wearables can calculate or estimate body position, movement velocity, distance traveled, and acceleration employing information provided by Global Navigation Satellite Systems (GNSS; such as the Global Positioning System; GPS) (Schutz and Chambaz, 1997; Cummins et al., 2013). With this technology, a good line-of-sight and high-sampling frequency are important for obtaining accurate data (Baranski and Strumillo, 2012; Cummins et al., 2013). Consequently, GNSS measurements do not function indoors or underwater and, moreover, their accuracy may be compromised in densely built-up areas. Inexpensive GPS systems are latent, a problem avoided by high-frequency sampling by professional systems.
In contrast, speed tracking appears to be accurate even with inexpensive systems. Position, velocity and distance measured at low-to-moderate velocities (<20 km·h−1) by such systems are also reliable, but acceleration data are prone to error and should be interpreted with caution (Cummins et al., 2013; Buchheit et al., 2014).
Accelerometers, which are commonly piezoelectric, piezoresistive, capacitive, or based on strain gauges (Kavanagh and Menz, 2008; Yang and Li, 2012), are used to quantify the distance an athlete covers during training, as well as to evaluate total sleep time and estimate sleep quality, thereby providing an estimate of the quality of sleep (Halson, 2014a). Distance is derived by most accelerometers from the number of steps taken and most count accurately at velocities >67 m·min−1 (1.12 m·s−1) (Feito et al., 2012), which, however, does not necessarily mean that they measure distance accurately. Accelerometers are reasonably reliable and valid for monitoring the quality and quantity of sleep in certain populations with an accuracy of 80% compared to polysomnography (Leeder et al., 2012;
Hausswirth et al., 2014). However, each accelerometer must be fitted securely to prevent motion artifacts (Yang and Hsu, 2010) and accelerometers often fail in detecting the state of wakefulness
in sleep periods. Therefore, other methods for the purpose of sleep monitoring are warranted (Sadeh, 2011).
Pulse oximetry exploits the fact that oxyhemoglobin and deoxyhemoglobin absorb near-infrared light maximally at different wavelengths to monitor the oxygen saturation of arterial blood continuously (Chan and Chan, 2013). These sensors are inexpensive, small and simple to use (Chan and Chan, 2013), but prone to potential error due to vasoconstriction, hypovolemia and artifacts caused by excessive movement (Chan and Chan, 2013; Windsor and Rodway, 2014), which limits their usefulness in cold environments and while exercising.
Parameters associated with HR can be monitored with chest belts, photoplethysmography, or various sensors incorporated into clothing. Although chest belts are widely used by athletes, they are experienced as uncomfortable (Buchheit, 2014; Spierer et al., 2015). Photoplethysmography involves a diode on the skin that emits red or near-infrared light that penetrates the underlying tissue and is then reflected back and detected by a photo sensor. This allows assessment of pulse rate with sufficient accuracy at rest, but the error of measurement can be dependent on the photosensitivity of the skin and increases during exercise due to motion artifacts (Schäfer and Vagedes, 2013; Spierer et al., 2015). Consequently, such data should be interpreted with caution. In the case of smart clothing, conducting or metal- coated fibers can be woven into the fabric or conducting inks can be printed onto the garment to monitor HR and associated parameters (Stoppa and Chiolerio, 2014). However, even though promising, only a few studies to date have evaluated the accuracy and reliability of smart clothing (Pandian et al., 2008; Curone et al., 2010) and more are warranted.
To monitor muscle activity by electromyography (EMG), electrodes woven into fabrics have been found to provide values similar to those obtained with traditional surface electrodes (Finni et al., 2007). The drawback of skin electrodes, however, is that
• they must be positioned accurately, preferably “in the midline of the muscle belly between the nearest innervation zone and the myotendinous junction furthest from this zone” (De Luca, 1997), since even small movements away from the innervation zone (e.g., 10% of the muscle length) reduce signal amplitude considerably (Belbasis and Fuss, 2015);
• they must have a tri-polar configuration to allow utilization of
“the double differential technique to eliminate the presence of crosstalk” (De Luca, 1997) between different muscles; and
• the signal-force relationship is non-linear and dependend on the number of motor units recruited in the vicinity of the electrode (De Luca, 1997).
Therefore, EMG fabrics designed to assess muscular activity are considered inaccurate. An alternative and promising approach involves incorporation of pressure sensors into compression garments (Belbasis and Fuss, 2015).
To access local muscle oxidative metabolism and to derive lactate thresholds non-invasively, devices which use near- infrared spectroscopy (NIRS) can be employed (Ferrari et al., 2004; Bellotti et al., 2013). These devices are efficient in terms of both time and cost (Bellotti et al., 2013), but are disturbed by
adipose tissue (Ferrari et al., 2004) and motion artifacts (Virtanen et al., 2011).
OVERVIEW OF COMMERCIALLY
AVAILABLE WEARABLE TECHNOLOGIES DESIGNED FOR USE BY ATHLETES
The present discussion here is based on information provided by the manufacturers on their websites. Since the list of available wearables is large and rapidly growing, those described here were chosen if the technology involved was indicated on the website and if they appeared to be the most advanced product of a given manufacturer for a specific purpose. Moreover, we focus solely on wearables that show promise for monitoring the training and health of athletes.
We summarize the wearables chosen (n = 36) in Table 1 [wrist-worn devices (n = 22)], Table 2 [clothing-based (n = 8)], and Table 3 [ear-worn (n = 4) and other devices (n = 3)], together with the parameters of interest which they monitor and the technology on which they are based. All of these wearables transmit the data they collect to an external device for further analysis and most provide immediate biofeedback to the athlete.
So far the accuracy, reliability, or validity of nine devices have been evaluated scientifically (for details please see Tables 1–3).
Wrist-Worn Devices
Most wrist-worn devices employ accelerometers (n = 16), gyroscopes (n = 3), GNSS (n = 8), (barometric) altimeters (n = 8), photoplethysmography (n = 8), additional chest belts (n = 8), sensors of skin temperature (n = 4), pulse oximeters (n = 2) and/or sensors of UV light (n = 1) to monitor duration of activity (n = 21), distance (n = 17), and velocity (n = 12) of an athlete’s locomotion, change in elevation (n = 10), environmental temperature (n = 1), altitude (n = 2), HR (n = 19), HRV (n = 2), neuromuscular fatigue (n = 1), UV radiation (n = 1), SpO2(n = 3), sleep quality and quantity (n = 14), and subjective markers (n = 1). However, HR recovery, humidity, hydration status, lactate thresholds and body temperature are not assessed. Furthermore, the only wearable that accesses subjective markers focuses on pain, but no other factors related to the training and health status of athletes.
The previous model of the Philips Actiwatch Spectrum Pro© (Philips Respironics, Murrysville, PA, USA) showed high accuracy to detect sleep compared to polysomnography, however, its ability to detect wakefulness is low (Marino et al., 2013).
The preceding model of the Withings Pulse Ox© (Withings SA, Issy-les-Moulineaux, France) overestimates sleep time with a validity of r = 0.92 compared to polysomnography (Ferguson et al., 2015).
The Polar V800 (Polar Electro, Kempele, Finland) is valid to detect RR intervals with an error of 0.09% and an intra-class correlation coefficient of >0.99 (Giles et al., 2016).
The preceding model of the Mio Alpha 2© gave HR values while walking, weight lifting, and biking that differed significantly from those obtained with a reference device and this model appears to be prone to motion artifacts (Spierer et al., 2015).
The Jawbone UP © (the model preceding the Jawbone UP3 we describe) was validated for measurement of total sleep time and time-point of awakening after sleep onset and showed good agreement with polysomnography (de Zambotti et al., 2015).
To the best of our knowledge all other devices have not been evaluated scientifically and, accordingly, the data they provide should be interpreted with considerable caution.
Devices Incorporated into Clothing
Specially designed (“smart”) clothing, ranging from shirts, shorts, hats/helmets to socks, can monitor several internal and external parameters of relevance to athletes. Most “smart”
clothing currently available utilizes accelerometers (n = 5), electrocardiography (n = 1), additional chest belts (n = 1), photoplethysmography (n = 2), and/or conducting fibers woven into the fabric (n = 2) to measure HR (n = 7), HR recovery (n = 2), HR variability (n = 2), neuromuscular fatigue (by EMG, n = 1), and lactate threshold (by NIRS, n = 1). To assess the external parameters duration (n = 3), distance (n = 2), velocity (n =2), and change in elevation (n = 1), “smart” clothes (with exception of the Zephyr BioHarnessTM 3) rely on the data transmitted by the companion smartphone. The BioHarnessTM 3 (Zephyr Technology Corp, Annapolis, USA) also aims to derive body temperature from the parameters it assesses (Zephyr Technology Corp, 2015). To date, environmental temperature, humidity, UV radiation, SpO2, hydration status, quantity and quality of sleep, and subjective factors have been neglected by designers of “smart”
clothing. Furthermore, no “smart” clothing presently available can provide immediate biofeedback to the athlete without the involvement of an external device.
The Hexoskin© vest (Carré Technologies Inc., Montreal, Québec Canada) provides reliable detection of an athlete’s HR when lying, sitting, standing or walking slowly (%CV < 0.79 ± 0.77; ICC > 0.96;Villar et al., 2015). However, measurement of the other parameters relevant to the training and health of athletes has not yet been validated, least of all when training.
The BioHarnessTM 3 has an acceptable level of validity and reliability for HR (r = ∼0.91, p < 0.01; %CV < 7.6), but increasing errors at higher velocity (Johnstone et al., 2012).
Measurement of HRR and HRV by this device has not been evaluated scientifically. Since, at least to our knowledge, no other form of “smart” clothing has yet been evaluated scientifically, the data they provide should be interpreted with due caution.
Ear-Worn Devices
Devices worn as an earplug (n = 3) or around the auricle (n = 1) use accelerometers (n = 3), pulse oximeters (n = 2), photoplethysmography (n = 2), temperature sensors (n = 1), gyroscopes, and magnetometers (n = 1) to assess duration (n = 3), distance covered by the athlete while training (n = 1), velocity (n = 1), HR (n = 4), HRV (n = 1), HRR (n = 1), SpO2(n = 3), and body temperature (n = 1). However, it should be noted that such devices measure variations in pulse rate rather than HRV directly (Schäfer and Vagedes, 2013). Parameters such as change in elevation, environmental temperature, humidity, altitude, neuromuscular fatigue, UV radiation, hydration status, lactate
TABLE1|Wristdevicesdesignedtomonitorparametersrelatedtothetrainingandhealthofathletes. DeviceTraining parameters monitored
HealthparametersmonitoredTechnologyemployedAdditionalcommentsandscientificevaluation PolarV800©(PolarElectro, 2015)D,L,V,El,Alt, HR,HRR,HRV, F
Slt,SlqGPS,HRchestbelt,additionalsensorsAdditionalSensorsfromPolarElectrorequired.Validto detectRRintervalswithanerrorof0.09%andanICC> 0.99(Gilesetal.,2016) MicrosoftBand2©(Microsoft, 2015)D,L,V,HRUV,Slt,SlqAccelerometer,ambientlightsensor,barometer,capacitive sensor,GPS,GSR,gyroscope,photoplethysmograph,skin temp.sensor,UVsensor amiigo©(Amiigo,2015)D,L,V,HR, (HRV)Oxy,Slt,SlqAccelerometer,pulseoximeter,temp.sensor FitbitSurge©(fitbitInc.,2015b)D,L,V,El,HRSlt,SlqAccelerometer,altimeter,digitalcompass,GPS,gyroscope, photoplethysmograph,ambientlightsensor SuuntoAmbit3Peak(HR) (Suunto,2015)D,L,V,El, Etemp,Alt,HRGPS,barometer,compass,HRchestbelt WithingsPulseOx©(Withings, 2015)D,L,El,HROxy,Slt,SlqAccelerometer,pulseoximeterTheprecedingmodeloverestimatessleeptime,buthada validityofr=0.92whencomparedtoagoldstandard (Fergusonetal.,2015) FitbitchargeHR©(fitbitInc., 2015a)D,L,El,HRSlt,SlqAccelerometer,altimeter,photoplethysmograph Garminvivoactive©(GarminLtd., 2015b)
D,L,V,HRSlt,SlqAccelerometer,GPS,GLONASS,HRchestbelt GarminvivosmartHR©(Garmin Ltd.,2015c)D,L,El,HRSlt,SlqAccelerometer,altimeter,photoplethysmograph Fitbitcharge©(fitbitInc.,2015b)D,L,ElSlt,SlqAccelerometer,altimeter GarminForerunner910XT© (GarminLtd.,2015a)D,L,V,El,HRAltimeter,GPS,HRchestbelt LGElectronicsLifebandTouch ActivityTracker©(LGElectronics, 2015)
D,L,V,El,HRAccelerometer,altimeter,HRchestbelt BasisPeak©(Basis,2015)D,HRSlt,SlqAccelerometer,photoplethysmograph,GSR,skintemp. sensor JawboneUP3©(Jawbone, 2015)D,HRSlt,SlqAccelerometer,bioimpedenceSensorFortotalsleeptime,sleepefficiency,andwakeaftersleep onset,theprecedingmodelshowedgoodagreementwith polyssomnographywithameandifference±SDof10.0± 20.5min;−1.9±4.2%and0.6±14.7min,respectively(de Zambottietal.,2015). (Continued)
TABLE1|Continued DeviceTraining parameters monitored
HealthparametersmonitoredTechnologyemployedAdditionalcommentsandscientificevaluation MioAlpha2©(MioGlobal,2015)D,L,V,HRAccelerometer,photoplethysmographItspreviousmodelshowedsignificantdifferencestoa referencedeviceformeasuringHRatwalking,biking (p<0.05)andweightlifting(p<0.01)(Spiereretal.,2015). Adidasmicoachsmartrun© (adidasPty.Ltd.,2014)D,L,V,HRGPS,photoplethysmograph Nike+Sportband©(+Shoe insert)(NikeInc.,2015a)D,L,V,HRHRchestbelt,piezoelectricembedinshoe Nike+SportwatchGPS©(Nike Inc.,2015b)
D,L,V,HRGPS,HRchestbelt MedisanaViFitconnectActivity Tracker©(Medisana,2015)D,LSlt,SlqAccelerometer PhilipsActiwatchSpectrum Pro©(Philips,2016)Slt,Slq,(Sub)Accelerometer,irradiancesensor,photopicilluminance sensor,PhotonFluxsensorSubjectivemarkerstoassesspainonly.Thesleepaccuracy oftheprecedingmodelwasr=0.86comparedto polysomnography(Marinoetal.,2013). PolarElectroLoop©(Polar Electro,2015)D,HRSlt,SlqAccelerometer,HRchestbelt SeraphimSenseAngelSensor© (SeraphimSenseLtd.,2014)D,HR(Oxy)Accelerometer,gyroscope,photoplethysmograph,skin temp.sensorBloodoxygensensorunderdevelopment Abbreviations:Alt,altitude;D,distancetraveled;El,changeinelevation;Etemp,environmentaltemperature;F,neuromuscularfatigue;GLONASS,GlobalNavigationSatelliteSystem;GPS,GeneralPositioningSystem;GSR,Galvanic SkinResponse;HR,heartrate;HRV,variabilityofheartrate;HRR,heartraterecovery;ICC,Intraclasscorrelationcoefficient;L,durationofexercise;Oxy,bloodoxygenation;Slq,sleepquality;Slt,sleeptime;Sub,subjectivemarkers; UV,ultravioletradiation;V,velocity;(),withrestrictions.
